Recent Interests

I am interested in all kinds of data mining techniques and applications, in particular, those of immediate business and scientific value. The following are my recent interests, starting from the most recent:

• Graph and network data: As social netwwork applications are becoming a trend in our daily life, it is increasingly important to extend database management and data mining techniques to graph and network data. Two new challenges are the steer data size and the relationship data. Many problems involving graph data such as (subgraph isomorphism) are inherently hard. My current interest involves extending database and data mining techniques to massive graph and network data to support advanced social network applications in a scalable manner.
• Data privacy: Data mining enables us to analyze corporate data to find ways to increase efficiency of an organization. Though the focus is on aggregate patterns pertaining a group of individuals, the fact that the raw data set is made available to data miners, which could be a third party, could present a threat to the privacy rights of individuals who contribute the records in the data set. Removing the identity of individuals from the data set is not sufficient to prevent such threats because the released data could be linked to other public databases to re-identifty the individuals through shared attributes (such as age, gender, zipcode). An interesting research problem is how to render the data so that sensitive information is masked and usefulness to data mining is retained.
• Data security: Data security differs from data privacy in that data sensitivity arises due to the confidentiality of data owned by organizations. Outsourcing data management to an untrusted cloud server is an example where the information in the outsourced data must not be disclosed to the cloud server. A promising technique, called searchable encryption, is outsourcing the data in some encrypted form, yet allowing the cloud server to be able to compute the result of user queries without decrypting the data.
• Mining massive datasets: Real life datasets are massive not only for a large number of records but also for a large number of features or dimensions. Web documents are one example whereeach document consists of words drawn from a English dictionray. Search of similar documents demands for novel techniques because a sequential scan of all documents scan is not scalable. Traditional indexing techniques cannot be easily applied to documents due to the lack of a fixed structure. Data stream is another example of massive data where only partial data is known and underlying patterns/trends will change as new data arrive.
• Mining biological datasets: Biological databases and problems present significant challenges: not only the size may be large, but also the structure is often very complex, data items are cross-referenced, interact with each other in domain-specific ways, and exist at multiple levels of abstraction. Compared to business problems, there is little user intuition to apply. My interest is to see how data mining techniques can be extended to address these new requirements.
• Actionable data mining: Data mining employs a notion of interestingness to extract interesting patterns. Most interestingness measures statistic based such as correlation. Such interest ingness measures do not directly address the economic return of business and help maximize the business goal under consideration. The notion of actionable data mining emphasizes modeling the goal of application right in the data mining problem statement. Examples are online recommendation, cross selling, and direct marketing.